Certara’s AI answer to medical writing challenges in pharma

For decades, the pharmaceutical industry, along with other highly regulated sectors, has grappled with the often-cumbersome process of medical writing. Writers must comply with strict guidelines from regulatory bodies such as the FDA and EMA when creating clinical trial reports and regulatory submissions. Failure to comply can result in delays or rejection.

The advent of transformer-based language models has promised to help streamline a variety of writing tasks, given their ability to generate human-like text and understand context. But they come with limitations. Most notably, they can “hallucinate” — generating generally plausible-sounding but wrong information. The off-the-shelf large language models also lack the domain-specific knowledge needed for regulatory compliance.

Introducing CoAuthor, an AI-powered medical writing tool

To address these challenges and accelerate medical writing, Certara has introduced CoAuthor, an advanced writing platform designed specifically for medical writers that combines generative AI with domain-specific knowledge.

The goal is to plug into medical writers’ workflow rather than completely overhaul established processes. Similar to how, say, GitHub Copilot helps programmers quickly churn out blocks of code, CoAuthor is more of an assistant than an autonomous writing system for writers.

“We’re not trying to say, ‘Here’s a stack of papers, now go write me a CSR, Mr. AI model.’ Instead, we’re trying to structure it in such a way that the AI is actually a colleague… to help get over that initial hump,” said Sean McGee, director of product at Certara.

Speeding up first drafts

A press release notes that medical writers can use the system to develop consistent and reproducible content, “improving the time to first draft by at least 30%.”

CoAuthor supports a variety of document types, including Clinical Study Reports (CSRs), Patient Narratives, Synopses, Protocols, and Toxicology Reports. The platform also incorporates Structured Content Authoring to support consistency across documents.

As alluded to earlier, CoAuthor is also designed to be focused and grounded. “We’re targeting specific sections within the document,” McGee said. “That way, it keeps the GPT focused on the topic at hand. It’s not wandering around pulling information from places that it shouldn’t be.” Another key feature of CoAuthor is its use of Retrieval-Augmented Generation (RAG) technology. “What’s cool about using RAG as part of that is it helps to narrow the focus down and say, ‘Okay, here’s the question we’re asking you, but I want you to constrain your answer to just this collection of information.'” This approach significantly cuts the risk of hallucinations while ensuring that the AI-generated content remains relevant and accurate.

Prioritizing precision and validation

The focus on precision was a core priority in an internal development of CoAuthor. “That was the biggest thing we had to overcome working internally with our medical writing team,” McGee said. “People would ask: ‘This looks good, but how do I know that it’s right?'”

Certara itself has a sizable stable of medical writers. “We have about 200 medical writers at Certara who have been… I can’t even begin to know the exact number of combined years of experience that is but it’s, I mean, obviously meaningful,” McGee noted. This extensive in-house expertise has been crucial in developing CoAuthor to meet the real needs of medical writers.

To address concerns about accuracy, Certara built various validation techniques into CoAuthor. “We have different ways built into the software to validate that and allow people to quickly check, ‘Yep, this is the document it came from, this is what it used,’ etc.”

Plugging into existing workflows

CoAuthor is designed to integrate seamlessly into existing workflows. “We’ve made everything that happens within the product accessible in Microsoft Word,” McGee noted. “It literally fits into your existing workflow in your existing space.”

The software also maintains a “human-in-the-loop” approach. As McGee explains, “Our human-in-the-loop model significantly reduces drafting time, while still enabling writers to use the generated content in the ways that they decide is best.”